"""
Byte Dream - Hugging Face API Examples
Complete examples for using Byte Dream with Hugging Face Hub
"""

# ============================================================================
# Example 1: Use Model from Hugging Face Hub (Download Locally)
# ============================================================================

print("=" * 70)
print("Example 1: Download and run model locally on CPU")
print("=" * 70)

from bytedream import ByteDreamGenerator

# Load model directly from Hugging Face
generator = ByteDreamGenerator(
    hf_repo_id="Enzo8930302/ByteDream",  # Replace with your repo
    config_path="config.yaml",
    device="cpu",
)

# Generate single image
image = generator.generate(
    prompt="A beautiful sunset over mountains, digital art, vibrant colors",
    negative_prompt="ugly, blurry, low quality, distorted",
    width=512,
    height=512,
    num_inference_steps=50,
    guidance_scale=7.5,
    seed=42,
)

image.save("example_1_output.png")
print("✓ Image saved to example_1_output.png\n")


# ============================================================================
# Example 2: Use Hugging Face Inference API (Cloud)
# ============================================================================

print("=" * 70)
print("Example 2: Use Hugging Face Inference API (No local computation)")
print("=" * 70)

from bytedream import ByteDreamHFClient

# Initialize client for API usage
api_client = ByteDreamHFClient(
    repo_id="Enzo8930302/ByteDream",
    token=None,  # Add your HF token here for private models
    use_api=True,  # Use cloud API instead of local inference
)

# Generate using cloud API
image_api = api_client.generate(
    prompt="Futuristic city with flying cars, cyberpunk style, night",
    negative_prompt="daylight, sunny, calm",
    width=512,
    height=512,
    num_inference_steps=50,
    guidance_scale=7.5,
)

image_api.save("example_2_api_output.png")
print("✓ Image generated via API and saved to example_2_api_output.png\n")


# ============================================================================
# Example 3: Batch Generation
# ============================================================================

print("=" * 70)
print("Example 3: Generate multiple images in batch")
print("=" * 70)

prompts = [
    "Majestic dragon flying over medieval castle, fantasy art",
    "Peaceful Japanese garden with cherry blossoms, serene",
    "Underwater coral reef with tropical fish, vibrant colors",
    "Mountain landscape at sunrise, dramatic lighting, epic",
]

# Generate all images
images = generator.generate_batch(
    prompts=prompts,
    negative_prompt="ugly, deformed, low quality",
    width=512,
    height=512,
    num_inference_steps=40,
    guidance_scale=7.5,
)

# Save all images
for i, img in enumerate(images):
    img.save(f"example_3_batch_{i+1}.png")
    print(f"  ✓ Saved example_3_batch_{i+1}.png")

print()


# ============================================================================
# Example 4: Upload Your Trained Model to Hugging Face
# ============================================================================

print("=" * 70)
print("Example 4: Upload trained model to Hugging Face Hub")
print("=" * 70)

# After training your model:
# python train.py --config config.yaml --train_data dataset

# Load your trained model
trained_generator = ByteDreamGenerator(
    model_path="./models/bytedream",
    config_path="config.yaml",
    device="cpu",
)

# Upload to Hugging Face
hf_token = "hf_xxxxxxxxxxxxx"  # Get from https://huggingface.co/settings/tokens

trained_generator.push_to_hub(
    repo_id="your_username/ByteDream",  # Replace with your username
    token=hf_token,
    private=False,  # Set True for private model
    commit_message="Upload Byte Dream model v1.0",
)

print("✓ Model uploaded to Hugging Face!\n")


# ============================================================================
# Example 5: Deploy to Hugging Face Spaces
# ============================================================================

print("=" * 70)
print("Example 5: Deploy interactive web app to Hugging Face Spaces")
print("=" * 70)

# Run this command in terminal:
# python deploy_to_spaces.py --repo_id your_username/ByteDream-Space

# Or programmatically:
from huggingface_hub import create_repo, HfApi

api = HfApi()

# Create space
create_repo(
    repo_id="your_username/ByteDream-Space",
    repo_type="space",
    sdk="gradio",
    token=hf_token,
    exist_ok=True,
)

# Upload files
api.upload_folder(
    folder_path=".",
    repo_id="your_username/ByteDream-Space",
    repo_type="space",
    token=hf_token,
    ignore_patterns=["*.git/*", "outputs/*", "logs/*"],
)

print("✓ Space deployed! Visit: https://huggingface.co/spaces/your_username/ByteDream-Space\n")


# ============================================================================
# Example 6: Advanced API Usage with Custom Parameters
# ============================================================================

print("=" * 70)
print("Example 6: Advanced generation with custom parameters")
print("=" * 70)

from bytedream import ByteDreamHFClient
import torch

client = ByteDreamHFClient(
    repo_id="Enzo8930302/ByteDream",
    use_api=False,  # Run locally
    device="cpu",
)

# Generate with different resolutions
resolutions = [(256, 256), (512, 512), (768, 768)]

for width, height in resolutions:
    print(f"\nGenerating {width}x{height} image...")
    
    img = client.generate(
        prompt="Abstract geometric patterns, colorful, modern art",
        width=width,
        height=height,
        num_inference_steps=30,  # Fewer steps for faster generation
        guidance_scale=9.0,  # Higher guidance for more detail
        seed=torch.randint(0, 1000000, (1,)).item(),
    )
    
    img.save(f"example_6_{width}x{height}.png")
    print(f"  ✓ Saved example_6_{width}x{height}.png")


# ============================================================================
# Example 7: Compare Local vs API Inference
# ============================================================================

print("\n" + "=" * 70)
print("Example 7: Compare local inference vs cloud API")
print("=" * 70)

import time

prompt = "Serene lake surrounded by pine trees, mountain reflection, sunset"

# Time local inference
print("\n⏱️ Testing LOCAL inference...")
start_local = time.time()
img_local = generator.generate(
    prompt=prompt,
    num_inference_steps=30,
    seed=123,
)
time_local = time.time() - start_local
img_local.save("comparison_local.png")
print(f"Local: {time_local:.2f}s")

# Time API inference
print("\n⏱️ Testing CLOUD API inference...")
start_api = time.time()
img_api = api_client.generate(
    prompt=prompt,
    num_inference_steps=30,
    seed=123,
)
time_api = time.time() - start_api
img_api.save("comparison_api.png")
print(f"API: {time_api:.2f}s")

print(f"\nSpeed comparison:")
print(f"  Local: {time_local:.2f}s (CPU)")
print(f"  API:   {time_api:.2f}s (Cloud GPU/CPU)")
print(f"  Winner: {'API' if time_api < time_local else 'Local'} 🏆")


print("\n" + "=" * 70)
print("All examples completed successfully! 🎉")
print("=" * 70)
print("\nNext steps:")
print("1. Train your own model: python train.py")
print("2. Upload to HF: generator.push_to_hub(repo_id='username/Model')")
print("3. Deploy to Spaces: python deploy_to_spaces.py")
print("4. Share with the community!")